Detailed urban object-based classifications from WorldView-2 … · 2014-07-03 · Detailed urban object-based classifications from ... 2007; Envi-zoom tutorial 2010) o Only based

#

Detailed urban object-based classifications from

WorldView-2 imagery and LiDAR data: supervised vs.

fuzzy rule-based

Commission 3

#

Alireza Hamedianfar¹

Assoc. Prof. Dr.Helmi Zulhaidi Mohd Shafri¹ʼ²

Prof. Dr. Shattri Mansor¹ʼ²

Assoc.Prof. Dr. Noordin Ahmad³

¹Department of Civil Engineering, Faculty of Engineering, Universiti Putra Malaysia(UPM)

²Geospatial Information Science Research Centre (GISRC), Faculty of Engineering,Universiti

Putra Malaysia (UPM)

³National Space Agency, Bangunan Komersil PJH, 62570 Putrajaya, Malaysia

[email protected]

[email protected]

#

Introduction

Content

�Introduction

� Objectives

� Study area

�Methodology

� Results and Discussions

�Conclusion

XXV International Federation of Surveyors

Congress, Kuala Lumpur, Malaysia, 16 – 21

June 2014

#

Introduction

Urbanization Phenomenon:

• Growth of built-up area and man-made

structures

• Rapid transformation of natural environment to

Impervious Surfaces (IS)

• IS distribution as a major contributor to

urbanization and environmental condition

(Arnold & Gibbons, 1996; Weng 2012).

• Roofs as the 25% coverage of urban areas

(Akbari et al. 2003)XXV International Federation of Surveyors


June 2014

#

IntroductionRoofing materials:

• Concrete tile

• Clay tile

• Metal

• Asbestos

• Polycarbonate

How roofing materials can affect

environmental condition and

quality?

• Pollution

• Health

• Heat/energy



June 2014

#

IntroductionPollution:Roof runoff as a source of water pollution

(Ballo et al. 2009)

Factors affecting surface water quality

(Göbel et al. 2007; Gikas and Tsihrintzis

2012):

• Roofing materials

Heavy metals such as zinc,

copper and cadmium have been

reported to be a source of

contamination for surface water (Van

Metre and Mahler 2003).

Other factors

• Conditions and slope of roofs

• Differences in usage of buildings (e.g.

residential or industrial)XXV International Federation of Surveyors


June 2014

#

Introduction

Health:

Asbestos roofs as contributor to Lung Cancer

• In 2004 , 107,000 deaths and 1,523,000 Disability Adjusted Life Years (DALYs)

(Frassy et al 2012)

• Asbestos roofs marketing and use was banned in European Union on January 1,

2005 (following the directive 76/769/CEE) (Frassy et al 2012).

• A consensus was achieved on banning of asbestos by Department of

Occupational Safety and Health (DOSH) on 28 March 2011 in Malaysia (Safitri et

al 2013).

• Complete ban in Malaysia is under negotiation and process (Safitri et al 2013).



June 2014

#

Introduction

� How we can get information about roof material types in large study area?

1- Field visit, but

• too labor intensive

• very time consuming for large study area.

2- Hyperspectral Remote sensing offer great potential to map surface materials, but

• Provides limited coverage

• too expensive

3- New generation of Very High Resolution (VHR) satellite imagery

• WorldView-2 image

• much more efficient and cost effective



June 2014

#

Classification Approach Advantage Disadvantage

spectral-basedImage analysis

� Sufficient for medium spatial multispectral image and hyperspectral imagery(Weng2007; Envi-zoom tutorial 2010)

o Only based on pixel-values (DNs).o Poor performance in handling VHR

imagery.o Not able to utilizespatial and textural

information (Blaschke and Strobl,2001; Myint et al. 2011).

Object-based image analysis (OBIA)

1-Supervised2-Rule-based

� Change the classification elementto image object.

� Utilization of all spatial, spectral,and texturalinformation.

� Able to provide advancedrule-setsto map different target (Myint etal. 2011; Bhaskaran et al. 2012;Hamedianfar and Shafri 2013).

o Often results in unclassfied andmixed objects when dealing withintra-urban targets (Pinho et al 2012,Hamedianfar and Shafri 2013)

Current trends with land cover classification



June 2014

#

• To identify optimal object-based rule-sets for roof material detection and other

urban features

• To provide cartographically pleasing outline of roof infrastructure by using

LiDAR and very high spatial resolution imagery of WorldView-2

• To compare the accuracy and efficiency of rule-based classifier and supervised

SVM classifier in utilizing of LiDAR and WorldView-2 image

Objectives



June 2014

#

Study area and data



June 2014

#

Data for this research rely on World-view2 (WV-2) imagery. WV-2 sensorincludes:

� Panchromatic channel(0.5m spatial resolution)

� Eight multispectral channels(1.8m spatial resolution)



June 2014

#

LiDAR (Light Detection and Ranging) data

LiDAR is an active remote sensing technology that measures distance with

reflected laser light (Jensen 2005)

3D view



June 2014

#

LiDAR (Light Detection and Ranging) data

2D view



June 2014

#

�Ancillary data : LiDAR products (DSM, DEM, nDSM¹)

nDSM image

WorldView-2 image

¹nDSM = Normalized Digital Surface Model



June 2014

#

Methodology

Land-cover classes

•Asbestos roof

•Concrete tile roof

•Metal roof

•Road

•Sidewalks

Segment Images

Merge Segments

Refine Segment

Compute Attributes

Define Features

classificationSupervised SVM

classificationclassificationFuzzy rule-based

classification

Export Features

Accuracy assessment

End

•Grass

•Trees

•Pond

•Swimming pool

•Shadow

RBF Kernel was used for SVM classifier. SVM

parameters ( C and γ) have been optimized

using parameter selection technique.(The cross

validation determined the optimal parameter C = 0.5 and

γ = 0.0078 with 5 fold function rate = 99 %)

Find Object

Extract Features

(ENVI-Zoom Tutorial (2010))XXV International Federation of Surveyors


June 2014

#

Supervised SVM object-based classification of LiDAR data and WorldView-2 image

Results and Discussions



June 2014

#

Note: Mr =metal roofs, Cr: concrete tile roofs, Ar: asbestos roofs, R= roads, S=sidewalks, G=grass, T=trees,

P=pond, Sp= swimming pool, Sh=shadow, Pa= producer accuracy, and Ua= user accuracy

Accuracy assessment of supervised SVM object-based classification

class Reference total Classified totalCorrect classified

Pa % Ua %

Mr 80 58 58 72.50 100Cr 58 57 47 81.03 82.46Ar 24 40 24 100 60R 61 43 41 67.21 95.35S 10 24 10 100 41.67G 52 52 51 98.08 98.08T 72 83 71 98.61 85.54P 33 38 33 100 86.84Sp 5 5 5 100 100Sh 39 34 29 74.36 85.29Overall Accuracy = 85.02% Kappa Coefficient = 0.82



June 2014

#

Rule-sets of object-based image analysis

(Adapted from ENVI-Zoom Tutorial (2010))



June 2014

#

Rule-sets of object-based image analysis

Land cover classes

Attribute MR CTR AR Roads SDW Trees Grass Lake SP SD

stdband6 <0.008 [-0.06,

0.01]

[-0.05, -

0.03]

nDSM >3.25 >2.41 [-0.28,

2.32]

>1.67 <9.5

maxband2 1.09

avgband2 >0.97 <1.12 <0.9 [1.51,

2.68]

maxband3 >1.32

avgband4 >0.89 [0.73,

1.49]

NDVI [0.66,

1.24]

[0.72,

0.85]

minband7 [0.72,

1.24]

[0.79,

0.82

[0.59,

0.85]

>0.857 >0.91

tx_entropy <0.18 >0.18

area >72.29 >161 >113.47 >202.6 >129.07 >10.8

1

avgband1 [0.87,1

.04]

>0.92

minband8 >0.35 >0.41

maindir <175.76 <0.92

elongation >1.33

tx_mean <0.06 [0.85,

0.97]

avgband3 >0.07

avgband5 <0.98

avgband8 <0.52 <0.82

hue <239.83

saturation >0.38

tx_range >0.03

avgband7 [0.39,

0.64]

minband1 <0.95

Note: MR= metal roofs, CTR= concrete tile roofs, AR= asbestos roofs, SP=

swimming pool, SDW =Sidewalk, and SD= shadows



June 2014

#

Rule-based object-based classification of LiDAR data and WorldView-2 image



June 2014

#

Note: Mr =metal roofs, Cr: concrete tile roofs, Ar: asbestos roofs, R= roads, S=sidewalks, G=grass, T=trees,

P=pond, Sp= swimming pool, Sh=shadow, Pa= producer accuracy, and Ua= user accuracy

Accuracy assessment of Rule-based object-based classification

class Reference total Classified totalCorrect classified

Pa % Ua %

Mr 80 80 78 97.50 97.50Cr 58 58 52 89.66 100Ar 24 24 24 100 100R 61 61 51 83.61 91.07S 10 10 10 100 100G 52 52 51 98.08 89.47T 72 72 71 98.61 92.21P 33 33 33 100 100Sp 5 5 5 100 100Sh 39 39 30 76.92 96.77

Overall Accuracy = 93.07% Kappa Coefficient = 0.92



June 2014

#

• Improvement of rule-based classification by nDSM and WorldView-2 data

fusion (93.02% overall accuracy)

• Higher accuracy by using the developed rule-sets from spectral, spatial,

texture and elevation information

• Effective reduction of misclassifications

• Increase the productivity and efficiency of OBIA

� Future study can be done to explore the transferability of established rule-

sets.

� Expand the work to larger study areas, and explore the performance of

different object-based classifiers

Conclusion & Recommendations



June 2014

#

References

Arnold Jr, C. L., & Gibbons, C. J. (1996). Impervious surface coverage: the emergence of a key environmental indicator. Journal

of the American Planning Association, 62(2), 243-258.

Ballo, S., M. Liu, L. Hou, and J. Chang. 2009. “Pollutants in Storm Water Runoff in Shanghai

(China): Implications for Management of Urban Runoff Pollution.” Progress in Natural Science

19 (7): 873–880.

Blaschke, T., Strobl, J., 2001 "What’s wrong with pixels? Some recent developments interfacing remote sensing and

GIS."GeoBIT/GIS6, pp. 12-17.

ENVI-Zoom. 2010. ENVI User Guide. Denver, CO: ITT.

F. Frassy, G. Candiani, P. Maianti, A. Marchesi, F. R. Nodari, M. Rusmini, C. Albonico, and M. Gianinetto, "Airborne remote

sensing for mapping asbestos roofs in aosta valley", in Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE

International, A. Moreira and Y.-L. Desnos, Eds., Proc IEEE, 7541-7544 (2012).

Göbel, P., C. Dierkes, and W. Coldewey. 2007. “Storm Water Runoff Concentration Matrix for

Urban Areas.” Journal of Contaminant Hydrology 91 (1): 26–42.

Hamedianfar, A., Shafri, H. Z. M., 2013. Development of fuzzy rule-based parameters for urban object-oriented classification

using very high resolution imagery. Geocarto International, pp. 1-25. doi: 10.1080/10106049.2012.760006

Hamedianfar, A., Shafri, H. Z. M., Mansor, S., & Ahmad, N. (2014). Improving detailed rule-based feature extraction of urban

areas from WorldView-2 image and lidar data. International Journal of Remote Sensing, 35(5), 1876-1899.

Myint, S W., Gober, P., Brazel A, Grossman-Clarke, S., Weng., Q, 2011. Per-pixel vs. object-based classification of urban land

cover extraction using high spatial resolution imagery. Remote Sensing of Environment, 115, pp. 1145-1161.



June 2014

#

References

Pinho, C. M. D., L. M. G. Fonseca, T. S. Korting, C. M. de Almeida, and H. J. H. Kux. 2012. “Land-Cover Classification of an Intra-

Urban Environment Using High-Resolution Images and Object-Based Image Analysis.” International Journal of Remote Sensing

33 (19): 5973–5995.

Safitri Zen, R. Ahamad, K. Gopal Rampal, and W. Omar. 2013 "Use of asbestos building materials in Malaysia: legislative

measures, the management, and recommendations for a ban on use". INT J OCCUP ENV HEAL. 19(3), 169-178

Jensen JR. 2005. Introductory digital image processing: a remote sensing perspective. Upper Saddle River (NJ): Prentice Hall.



June 2014

#

Acknowledgement

• Financial assistance from Peninsula Malaysia

Land Surveyors Board (LJT).

• Financial support from the Ministry of Higher

Education(MOHE) Malaysia.



June 2014

#

Thank you for your attention!

¹Department of Civil Engineering, Faculty of Engineering, Universiti Putra

Malaysia (UPM)

²Geospatial Information Science Research Centre (GISRC), Faculty of Engineering,

Universiti Putra Malaysia (UPM)

Email: [email protected]



June 2014

Documents

Detailed urban object-based classifications from WorldView-2 … · 2014-07-03 · Detailed urban object-based classifications from ... 2007; Envi-zoom tutorial 2010) o Only based